Data-Driven Deep Learning Neural Networks for Predicting the Number of Individuals Infected by COVID-19 Omicron...

Area of Science:

• Epidemiology
• Computational Biology
• Data Science

Background:

• Real-time epidemic prediction is vital for public health interventions.
• Mathematical models like SIR are essential tools for disease forecasting.
• Data-driven deep learning offers advanced methods for parameter identification in epidemic models.

Purpose of the Study:

• To develop and evaluate predictive models for COVID-19 Omicron variant infections.
• To compare the performance of traditional mathematical models with a novel time-series neural network approach.
• To assess model accuracy across countries with diverse public health mitigation strategies.

Main Methods:

• Reduced the SIR model to logistic differential equations, creating constant, rational, and birational models.
• Developed a time-series model utilizing neural networks for infection prediction.
• Introduced a logistics-informed neural network algorithm to determine analytical solutions from data.
• Validated model accuracy using error metrics on Omicron variant data from Portugal, Italy, and China.

Main Results:

• Constant models showed poor predictive accuracy for daily and cumulative Omicron infections.
• Rational and birational models accurately predicted cumulative infections in countries with strict mitigation but struggled with daily infections and partial mitigation.
• The novel time-series model demonstrated versatility and accuracy in predicting both daily and cumulative infections, regardless of mitigation stringency.

Conclusions:

• Traditional mathematical models have limitations in predicting COVID-19 Omicron variant spread, especially with partial mitigation measures.
• A time-series neural network model offers a robust and versatile solution for real-time epidemic forecasting.
• Accurate prediction of infectious disease dynamics is achievable with advanced data-driven approaches, aiding public health preparedness.